Apple's Machine Learning Research division, the Core ML and Foundation Models teams, and the AIML Residency program together run the most selective, least-publicized early-career AI pipeline in Cupertino — and the Class of 2026 landing inside it is making a structurally different bet from the cohort joining OpenAI or Anthropic. The new-grad AI slots inside Apple's 20,000-worker US hiring plan, announced February 2025, are few. That scarcity is the point.
What Happened
Apple's on-device AI hiring this cycle runs through three distinct entry points.
The first is the AIML Residency — a one-year program for graduates with advanced degrees, now in its sixth cohort. Base pay for the 2026 cohort is published in Apple's job filings: the range sits between $143,100 and $214,500 depending on experience and location. Residents join an Apple product team for the year and ship work that lands in the hands of the one billion active iPhone users Apple disclosed in its most recent earnings. The Residency accepts candidates with backgrounds in machine learning, linguistics, cognitive science, and fair and responsible AI — a deliberately cross-disciplinary intake that reflects where Apple's on-device intelligence problems actually sit. Unlike an OpenAI fellowship or an Anthropic Fellows track, the Residency is not structured around research publication. The output measure is product impact, not papers.
The second entry point is the Apple Scholars in AIML PhD fellowship — now in its seventh year, with more than 120 scholars supported to date. Apple does not accept direct PhD applications. Universities nominate. The 2026 cohort includes Fengrui Tian from the University of Pennsylvania, Ephraim Linder from Boston University, and Kyra Wilson from the University of Wisconsin, among others confirmed by their respective institutions. Research areas for the 2026 fellowship class span privacy-preserving machine learning, ML algorithms and architectures, interactive ML and agents, and data-centric AI — a list that maps precisely onto the problems Apple's Core ML and Foundation Models teams are shipping against right now. Scholars receive funding, mentorship from Apple researchers, and an internship. A portion of those interns convert to full-time roles inside Cupertino's ML division.
The third entry point is the direct new-grad ICT2 hire into the on-device ML engineering org. Apple does not publish its entry-level headcount for this role tier. The compensation structure is public on Levels.fyi: ICT2 ML engineers report a base salary in the $141,000 range, with total first-year compensation around $180,000 at the median. RSU grants at ICT2 land in the $80,000 to $150,000 range over four years at standard offer, negotiable toward $150,000 to $220,000 with a competing offer from Google, Meta, or a well-funded lab. Signing bonuses of $20,000 to $50,000 are available. Apple's RSU structure vests 25 percent per year, semi-annually, with no cliff — a meaningful difference from OpenAI's Profit Participation Units and from Anthropic's 4-year vest with a one-year cliff.
The interview pipeline that leads to an ICT2 or Residency offer is structurally heavier than what most candidates expect from Apple. For ML engineering roles, the process runs a recruiter screen, two phone rounds, and an on-site loop of five to eight interviews. Median time-to-offer sits around 21 days from first contact, per public Glassdoor and IGotAnOffer reporting — faster than OpenAI's publicly cited 32-day average. The technical depth, however, is specific to Apple's constraint set. Interview questions tested at the on-site level include: tradeoffs between on-device inference and server-side inference for Siri-class latency; training frameworks for face recognition models running on iPhone silicon; and the impact of model quantization on accuracy when operating within the power envelope of the Apple Neural Engine. These are not generic ML systems questions. They are Apple-shaped questions, and candidates who have not studied the constraints of the Neural Engine, the M-series chip family, and the Core ML inference stack fail on specifics that are not covered by the standard LeetCode + ML fundamentals preparation path that suffices at a frontier lab.
Why It Matters
The comparison that matters for the 2026 graduate choosing between Apple and a frontier lab comes down to three variables: compensation structure, research posture, and career trajectory.
On compensation, OpenAI's entry-level research positions pay total comp of approximately $180,000 to $220,000 in year one at the L2 equivalent level, per Levels.fyi public submissions. Anthropic's Fellows track — the primary new-grad research entry point — paid $3,850 per week for a four-month term, with a reported 25 to 50 percent conversion rate to full-time. Apple's AIML Residency base of $143,100 to $214,500 is broadly competitive with those ranges. The structural difference is that Apple's on-device engineers participate in Apple's employee stock purchase plan and standard RSU program against a public company equity — AAPL traded between $190 and $260 for most of 2025. A new grad entering on an RSU grant of $120,000 over four years is holding equity in a $3 trillion market cap company, not a pre-liquidity valuation that requires an IPO or acquisition event to crystallize. That is a fundamentally different risk-return profile from the PPU or RSU packages that Anthropic and OpenAI offer.
On research posture, Apple's ML hiring is emphatically not publication-first. The on-device ML team's mandate — running inference on the Neural Engine, optimizing 3-billion-parameter on-device models for multilingual and multimodal tasks across iPhone, Mac, and Vision Pro — is a systems and applied research problem, not a frontier scaling problem. Graduates who want to publish at NeurIPS and ICML will find Apple's research culture slow and proprietary. Graduates who want to see their work run on a billion devices inside 12 months of joining, operated within a privacy constraint that no cloud-forward lab applies to its inference stack, are choosing a fundamentally different kind of impact. Apple's Core OS Machine Learning and Differential Privacy team — a publicly posted role on LinkedIn — represents the far edge of that constraint: engineers building the local differentially private algorithms that allow Apple to learn statistical trends without user-level data ever leaving the device.
Apple's privacy-first AI architecture is not a brand position. It is a technical constraint that defines the hiring profile. AppleInsider reported in May 2026 that privacy and data security will remain the central organizing principle of Apple's 2026 AI push, with the company's 2026 AI strategy explicitly prioritizing on-device models powered by the M5 chip family's neural accelerators over server-side inference. Analysts at several buy-side firms cited by AppleMagazine estimate Apple will expand its AI teams by more than 20 percent through 2026 while holding total headcount roughly flat — which means the growth is concentrated, not distributed. The engineering roles being created inside that 20 percent expansion are specific: on-device inference optimization, differential privacy at scale, multimodal model compression, and the Neural Engine software stack that sits between Core ML and the silicon. The AIML Residency and the ICT2 ML track are the primary graduate entry points into that expansion.
What's Next
Three signals to track through the second half of 2026.
The intern conversion window. Apple's standard recruiting calendar converts summer 2026 interns to full-time ICT2 offers in September and October. The volume and velocity of that conversion round — relative to 2025 — will be the clearest leading indicator of whether Apple's on-device AI team is expanding at the pace its M5-cycle roadmap implies. Apple does not publish intern conversion data. The signal will appear first in LinkedIn profile updates from CMU, Stanford, MIT, and UIUC PhD students who interned in Apple's Core ML and Foundation Models groups this summer.
The Foundation Models team's publication posture. Apple's Foundation Models group, restructured in late 2025 to sit independently of the Siri and robotics orgs, has increased its presence at ML conferences since the Apple Intelligence launch. The 2026 WWDC period — running through June — will signal whether Apple is willing to put more technical detail around its on-device LLMs into the public record. Graduate candidates making decisions in August will read that signal: more published specifics means the research posture is shifting, and the career trade-off between Apple and a frontier lab on research visibility narrows.
Competing offer pressure from NVIDIA and Google. NVIDIA's NCG rotation, with its CUDA-depth credential and publicly cited 85 to 90 percent intern-to-offer rate, and Google DeepMind's US graduate research program are the two most direct competitors for the profile Apple recruits. Both offer higher headline total compensation at the new-grad level — NVIDIA's IC1 ML engineers land between $335,000 and $430,000 in year one, per Levels.fyi. Apple's ICT2 at around $180,000 median is not the same number. It persists because the candidates who accept it are not optimizing on year-one cash or on research publication velocity. They are betting that building AI that runs privately on a billion devices is a harder and more durable engineering credential than building AI that runs on a server. For the 2026 class that has internalized the privacy-preserving ML research agenda — the Apple Scholars fellowship covers it explicitly — that bet is legible and deliberate.
Apple's graduate AI pipeline is quiet by design. The company does not run campus recruiting events with the visibility of Google or the intensity of OpenAI. It does not publish intern conversion rates, cohort sizes, or team headcounts. What it publishes is its research — differential privacy at scale, on-device foundation models, privacy-preserving aggregate trend analysis — and the candidates who have read it know exactly what they are applying to join.
Compensation data sourced from Levels.fyi public submissions, last updated May 16, 2026, and Candor ICT2 offer aggregator. Apple AIML Residency base pay range from publicly filed Apple job descriptions on jobs.apple.com. 2026 Apple Scholars in AIML cohort members confirmed by University of Pennsylvania, Boston University, and University of Wisconsin public announcements. Apple's US hiring plan from company press release, February 2025, cross-referenced by Fortune and Yahoo Finance reporting. Apple AI team headcount growth estimate from analyst commentary cited by AppleMagazine. Interview process timing from IGotAnOffer and Glassdoor public reporting. Apple's 2026 AI privacy strategy from AppleInsider, May 17, 2026. NVIDIA ICT1 compensation from Levels.fyi public submissions, cited in ENTRA US Bureau briefing, May 2026. Fellowship program history and 2026 research areas from Apple Machine Learning Research public announcement.